// TestError makes sure that we handle commands that error correctly.
func TestError(t *testing.T) {
t.Parallel()
inRepo := "TestError_in"
require.NoError(t, btrfs.Init(inRepo))
outPrefix := "TestError_out"
// Create the Pachfile
require.NoError(t, btrfs.WriteFile(path.Join(inRepo, "master", "pipeline", "error"), []byte(`
image ubuntu
run touch /out/foo
run cp /in/foo /out/bar
`)))
// Last line should fail here.
// Commit to the inRepo
require.NoError(t, btrfs.Commit(inRepo, "commit", "master"))
err := RunPipelines("pipeline", inRepo, outPrefix, "commit", "master", "0-1", etcache.NewCache())
require.Error(t, err, "Running pipeline should error.")
// Check that foo exists
exists, err := btrfs.FileExists(path.Join(outPrefix, "error", "commit-0", "foo"))
require.NoError(t, err)
require.True(t, exists, "File foo should exist.")
// Check that commit doesn't exist
exists, err = btrfs.FileExists(path.Join(outPrefix, "error", "commit"))
require.NoError(t, err)
require.False(t, exists, "Commit \"commit\" should not get created when a command fails.")
}
func (r *Runner) makeOutRepo(pipeline string) error {
if err := btrfs.Ensure(path.Join(r.outPrefix, pipeline)); err != nil {
return err
}
exists, err := btrfs.FileExists(path.Join(r.outPrefix, pipeline, r.branch))
if err != nil {
return err
}
if !exists {
// The branch doesn't exist, we need to create it We'll make our branch
// have the same parent as the commit we're running off of if that
// parent exists in the pipelines outRepo. This lets us carry over past
// computation results when a new branch is created rather than having
// to start from scratch.
parent := btrfs.GetMeta(path.Join(r.inRepo, r.commit), "parent")
if parent != "" {
exists, err := btrfs.FileExists(path.Join(r.outPrefix, pipeline, parent))
if err != nil {
return err
}
if !exists {
parent = ""
}
}
if err := btrfs.Branch(path.Join(r.outPrefix, pipeline), parent, r.branch); err != nil {
return err
}
}
// The branch exists, so we're ready to return
return nil
}
func TestInputOutput(t *testing.T) {
t.Parallel()
// create the in repo
pipeline := newTestPipeline(t, "inputOutput", "commit", "master", "0-1", true)
// add data to it
err := btrfs.WriteFile(path.Join(pipeline.inRepo, "master", "data", "foo"), []byte("foo"))
require.NoError(t, err)
// commit data
err = btrfs.Commit(pipeline.inRepo, "commit", "master")
require.NoError(t, err)
pachfile := `
image ubuntu
input data
run cp /in/data/foo /out/foo
`
err = pipeline.runPachFile(strings.NewReader(pachfile))
require.NoError(t, err)
exists, err := btrfs.FileExists(path.Join(pipeline.outRepo, "commit-0", "foo"))
require.NoError(t, err)
require.True(t, exists, "File `foo` doesn't exist when it should.")
}
// TestScrape tests a the scraper pipeline
func TestScrape(t *testing.T) {
// TODO(any): what?? wget is not found in the container if parallel is set
//t.Parallel()
pipeline := newTestPipeline(t, "scrape", "commit", "master", "0-1", true)
// Create a url to scrape
require.NoError(t, btrfs.WriteFile(path.Join(pipeline.inRepo, "master", "urls", "1"), []byte("pachyderm.io")))
// Commit the data
require.NoError(t, btrfs.Commit(pipeline.inRepo, "commit", "master"))
// Create a pipeline to run
pachfile := `
image pachyderm/scraper
input urls
run cat /in/urls/* | xargs wget -P /out
`
err := pipeline.runPachFile(strings.NewReader(pachfile))
exists, err := btrfs.FileExists(path.Join(pipeline.outRepo, "commit", "index.html"))
require.NoError(t, err)
require.True(t, exists, "pachyderm.io should exist")
}
func (p *pipeline) fail() error {
exists, err := btrfs.FileExists(path.Join(p.outRepo, p.commit+"-fail"))
if err != nil {
return err
}
if exists {
return nil
}
return btrfs.DanglingCommit(p.outRepo, p.commit+"-fail", p.branch)
}
// finish makes the final commit for the pipeline
func (p *pipeline) finish() error {
exists, err := btrfs.FileExists(path.Join(p.outRepo, p.commit))
if err != nil {
return err
}
if exists {
return nil
}
return btrfs.Commit(p.outRepo, p.commit, p.branch)
}
func TestEcho(t *testing.T) {
t.Parallel()
pipeline := newTestPipeline(t, "echo", "commit", "master", "0-1", true)
pachfile := `
image ubuntu
run echo foo >/out/foo
run echo foo >/out/bar
`
err := pipeline.runPachFile(strings.NewReader(pachfile))
require.NoError(t, err)
exists, err := btrfs.FileExists(path.Join(pipeline.outRepo, "commit-0", "foo"))
require.NoError(t, err)
require.True(t, exists, "File `foo` doesn't exist when it should.")
exists, err = btrfs.FileExists(path.Join(pipeline.outRepo, "commit-1", "bar"))
require.NoError(t, err)
require.True(t, exists, "File `bar` doesn't exist when it should.")
}
// Run runs a command in the container, it assumes that `branch` has already
// been created.
// Notice that any failure in this function leads to the branch having
// uncommitted dirty changes. This state needs to be cleaned up before the
// pipeline is rerun. The reason we don't do it here is that even if we try our
// best the process crashing at the wrong time could still leave it in an
// inconsistent state.
func (p *pipeline) run(cmd []string) error {
// this function always increments counter
defer func() { p.counter++ }()
// Check if the commit already exists
exists, err := btrfs.FileExists(path.Join(p.outRepo, p.runCommit()))
if err != nil {
return err
}
// if the commit exists there's no work to be done
if exists {
return nil
}
// Set the command
p.config.Config.Cmd = []string{"sh"}
//p.config.Config.Volumes["/out"] = emptyStruct()
// Map the out directory in as a bind
hostPath := btrfs.HostPath(path.Join(p.outRepo, p.branch))
bind := fmt.Sprintf("%s:/out", hostPath)
p.config.HostConfig.Binds = append(p.config.HostConfig.Binds, bind)
log.Print(p.config.HostConfig.Binds)
// Make sure this bind is only visible for the duration of run
defer func() { p.config.HostConfig.Binds = p.config.HostConfig.Binds[:len(p.config.HostConfig.Binds)-1] }()
// Start the container
p.container, err = startContainer(p.config)
if err != nil {
return err
}
if err := pipeToStdin(p.container, strings.NewReader(strings.Join(cmd, " ")+"\n")); err != nil {
return err
}
// Create a place to put the logs
f, err := btrfs.CreateAll(path.Join(p.outRepo, p.branch, ".log"))
if err != nil {
return err
}
defer f.Close()
// Copy the logs from the container in to the file.
if err = containerLogs(p.container, f); err != nil {
return err
}
// Wait for the command to finish:
exit, err := waitContainer(p.container)
if err != nil {
return err
}
if exit != 0 {
// The command errored
return fmt.Errorf("Command:\n\t%s\nhad exit code: %d.\n",
strings.Join(cmd, " "), exit)
}
return btrfs.Commit(p.outRepo, p.runCommit(), p.branch)
}
// TestWrap tests a simple pipeline that uses line wrapping in it's Pachfile
func TestWrap(t *testing.T) {
t.Parallel()
outRepo := "TestWrap_out"
require.NoError(t, btrfs.Init(outRepo))
pipeline := newPipeline("output", "", outRepo, "commit", "master", "0-1", "", etcache.NewCache())
pachfile := `
image ubuntu
# touch foo and bar
run touch /out/foo \
/out/bar
`
err := pipeline.runPachFile(strings.NewReader(pachfile))
require.NoError(t, err)
exists, err := btrfs.FileExists(path.Join(outRepo, "commit", "foo"))
require.NoError(t, err)
require.True(t, exists, "File `foo` doesn't exist when it should.")
exists, err = btrfs.FileExists(path.Join(outRepo, "commit", "bar"))
require.NoError(t, err)
require.True(t, exists, "File `bar` doesn't exist when it should.")
}
func (p *pipeline) finished() (bool, error) {
return btrfs.FileExists(path.Join(p.outRepo, p.commit))
}
func checkExists(t *testing.T, name string) {
exists, err := btrfs.FileExists(name)
require.NoError(t, err)
require.True(t, exists)
}